Developing robust Data Mining (DM) methods for human activity recognition based on real-time smart-home or smart-phone sensor data is a challenging task. Problems occur with correctly classifying peoples' actions caused by observations which do not conform to an expected pattern. This is due to variations in how they execute activities and the duration of these events. This is an important issue to study as DM applications can be in the area of monitoring of daily activities for the elderly and identifying anomalous behaviour so that alerts can be sent to relatives or caregivers.
Compounding this problem is that identification of these activities need to occur in the presence of unforeseen changes in the statistical properties of the data; an effect known as concept drift. DM methods therefore need to also automatically adapt themselves to account for concept drift thus maintaining their efficacy.
This seminar covers some recent work on human activity recognition based on smart-home sensor data and proposes a new DM framework to model this task. The objective of this work is to ultimately identify the possible existence of concept drift and suggest how DM methods for this task could be improved upon.
Last modified: Tuesday, 06-Mar-2018 11:22:15 NZDT
This page is maintained by the seminar list administrator.